文章基本信息

标题：Neural ODEs and differential flatness for total least squares parameter estimation
本地全文：下载
作者：Aike Aline Tappe ; Moritz Schulze ; René Schenkendorf 等
期刊名称：IFAC PapersOnLine
印刷版ISSN：2405-8963
出版年度：2022
卷号：55
期号：20
页码：421-426
DOI：10.1016/j.ifacol.2022.09.131
语种：English
出版社：Elsevier
摘要：AbstractIn (bio)chemical process engineering, first-principles process models have played a central role for some time in better understanding, monitoring, and controlling these complex processes. Dynamic process models have become even more critical in the context of Industry 4.0 and the use of digital twins in the last decade. However, the quality and the technology readiness level of digital process models depend crucially on the reliability of the model predictions. In addition to a suitable model structure/hypothesis, the model parameters of the implemented kinetics are of paramount importance. The accuracy of the parameter estimation, in turn, depends on the quantity and quality of the data as well as on the employed parameter identification solving strategies, where ordinary least squares concepts are still the standard. We propose a novel parameter identification concept that combines systems theory and machine learning principles. The parameter identification problem is formulated as a total least squares optimization problem that uses neural ordinary differential equations for surrogate modeling and recalculates the model control inputs with the algebraic differential flatness framework for model inversion. The usefulness of the proposed concept for more precise kinetic parameters is demonstrated with a simulation study of an enzyme-catalyzed biochemical process, where the total least squares approach leads to lower parameter uncertainties compared to the standard concept based on ordinary least squares using the same amount of data.
关键词：Keywordstotal least squaresneural ordinary differential equationsdifferential flatnessmodel inversionkinetic parameter estimationparameter uncertainties